Identification Of Generalized Myasthenia Gravis (MG) & Antibody Status Using Rules-Based Natural Language Processing (NLP) Applied To Neurologist Clinical Notes

This poster was originally presented at MGFA AANEM 2023 on November 1-4 in Phoenix, AZ.

Authors: Jonathan Darer, MD, MPH, Jacqueline Pesa, PhD, MPH, Zia Choudhry, MD, Alberto E. Batista, PharmD, MS, Purva Parab, PhD, Xiaoyun Yang, MS, and Raghav Govindarajan, MD

Affiliations: Health Analytics, LLC, Clarksville, MD; Janssen Scientific Affairs, LLC, Titusville, NJ, USA; Janssen Pharmaceutical Companies of Johnson & Johnson, and Hospital Sisters Health Systems Medical Group

Introduction

Myasthenia gravis (MG) is a rare chronic autoantibody-mediated neuromuscular disorder affecting approximately 60,000 individuals in the United States. The objective of this study is to assess the feasibility of identifying meaningful MG clinical subtypes from neurologist clinical progress notes.

Methods

• This was a retrospective, cross-sectional analysis using the Amplity AnswerY™ (formerly known as Amplity Insights) deidentified, unstructured medical transcription records of neurologists in the U.S. between 2017 and 2022, and covering 150,000 physicians from over 40,000 clinics and hospitals
• A rules-based NLP model, developed using spaCy (an open-source NLP library), was used to create an analyzable dataset
• The performance of the NLP model was assessed against manual annotations using Prodigy

Results

• A total of 3085 patients with MG were identified; most had unspecified clinical subtype (80.2%), followed by generalized MG (gMG) (10.6%), ocular MG (8.2%), and bulbar MG (2.4%) (Note: patients could be labeled with more than 1 subtype in different notes)
• Most patients did not have a documented MG antibody status (76.0%)

• For patients with unspecified clinical subtype, symptoms provided potential insight in 63% of patients, as evidenced by presence of symptoms indicative of ocular MG or gMG

Conclusions

• AnswerY was able to identify MG clinical subtypes for >20% of patients
• Using identified MG symptoms, it was possible to infer a clinical subtype for an additional 63% of patients with unspecified MG
• Antibody status was documented in almost 25% of patients with MG in the neurologist clinical progress notes
• MG treatment is increasingly being determined by clinical subtype and antibody status. Symptom documentation can identify patients with gMG and more severe disease manifestations

DOWNLOAD POSTER

Learn how else our Amplity AI-powered data does more.